?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / net / ipv6 / udp.c
blob1979922bcf673b271e711b2b9ca7ca7c95bceb2f
1 /*
2 * UDP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on linux/ipv4/udp.c
9 *
10 * Fixes:
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
39
40 #include <net/addrconf.h>
41 #include <net/ndisc.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_route.h>
45 #include <net/raw.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/xfrm.h>
49 #include <net/inet_hashtables.h>
50 #include <net/inet6_hashtables.h>
51 #include <net/busy_poll.h>
52 #include <net/sock_reuseport.h>
53
54 #include <linux/proc_fs.h>
55 #include <linux/seq_file.h>
56 #include <trace/events/skb.h>
57 #include "udp_impl.h"
58
59 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
60 {
61 #if defined(CONFIG_NET_L3_MASTER_DEV)
62 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
63 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
64 return true;
65 #endif
66 return false;
67 }
68
69 static u32 udp6_ehashfn(const struct net *net,
70 const struct in6_addr *laddr,
71 const u16 lport,
72 const struct in6_addr *faddr,
73 const __be16 fport)
74 {
75 static u32 udp6_ehash_secret __read_mostly;
76 static u32 udp_ipv6_hash_secret __read_mostly;
77
78 u32 lhash, fhash;
79
80 net_get_random_once(&udp6_ehash_secret,
81 sizeof(udp6_ehash_secret));
82 net_get_random_once(&udp_ipv6_hash_secret,
83 sizeof(udp_ipv6_hash_secret));
84
85 lhash = (__force u32)laddr->s6_addr32[3];
86 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
87
88 return __inet6_ehashfn(lhash, lport, fhash, fport,
89 udp_ipv6_hash_secret + net_hash_mix(net));
90 }
91
92 int udp_v6_get_port(struct sock *sk, unsigned short snum)
93 {
94 unsigned int hash2_nulladdr =
95 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
96 unsigned int hash2_partial =
97 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
98
99 /* precompute partial secondary hash */
100 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
101 return udp_lib_get_port(sk, snum, hash2_nulladdr);
102 }
103
104 static void udp_v6_rehash(struct sock *sk)
105 {
106 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
107 &sk->sk_v6_rcv_saddr,
108 inet_sk(sk)->inet_num);
109
110 udp_lib_rehash(sk, new_hash);
111 }
112
113 static int compute_score(struct sock *sk, struct net *net,
114 const struct in6_addr *saddr, __be16 sport,
115 const struct in6_addr *daddr, unsigned short hnum,
116 int dif, int sdif, bool exact_dif)
117 {
118 int score;
119 struct inet_sock *inet;
120
121 if (!net_eq(sock_net(sk), net) ||
122 udp_sk(sk)->udp_port_hash != hnum ||
123 sk->sk_family != PF_INET6)
124 return -1;
125
126 score = 0;
127 inet = inet_sk(sk);
128
129 if (inet->inet_dport) {
130 if (inet->inet_dport != sport)
131 return -1;
132 score++;
133 }
134
135 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
136 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
137 return -1;
138 score++;
139 }
140
141 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
142 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
143 return -1;
144 score++;
145 }
146
147 if (sk->sk_bound_dev_if || exact_dif) {
148 bool dev_match = (sk->sk_bound_dev_if == dif ||
149 sk->sk_bound_dev_if == sdif);
150
151 if (!dev_match)
152 return -1;
153 if (sk->sk_bound_dev_if)
154 score++;
155 }
156
157 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
158 score++;
159
160 return score;
161 }
162
163 /* called with rcu_read_lock() */
164 static struct sock *udp6_lib_lookup2(struct net *net,
165 const struct in6_addr *saddr, __be16 sport,
166 const struct in6_addr *daddr, unsigned int hnum,
167 int dif, int sdif, bool exact_dif,
168 struct udp_hslot *hslot2, struct sk_buff *skb)
169 {
170 struct sock *sk, *result;
171 int score, badness;
172 u32 hash = 0;
173
174 result = NULL;
175 badness = -1;
176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
177 score = compute_score(sk, net, saddr, sport,
178 daddr, hnum, dif, sdif, exact_dif);
179 if (score > badness) {
180 if (sk->sk_reuseport &&
181 sk->sk_state != TCP_ESTABLISHED) {
182 hash = udp6_ehashfn(net, daddr, hnum,
183 saddr, sport);
184
185 result = reuseport_select_sock(sk, hash, skb,
186 sizeof(struct udphdr));
187 if (result && !reuseport_has_conns(sk, false))
188 return result;
189 }
190 result = sk;
191 badness = score;
192 }
193 }
194 return result;
195 }
196
197 /* rcu_read_lock() must be held */
198 struct sock *__udp6_lib_lookup(struct net *net,
199 const struct in6_addr *saddr, __be16 sport,
200 const struct in6_addr *daddr, __be16 dport,
201 int dif, int sdif, struct udp_table *udptable,
202 struct sk_buff *skb)
203 {
204 struct sock *sk, *result;
205 unsigned short hnum = ntohs(dport);
206 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
207 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
208 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
209 int score, badness;
210 u32 hash = 0;
211
212 if (hslot->count > 10) {
213 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
214 slot2 = hash2 & udptable->mask;
215 hslot2 = &udptable->hash2[slot2];
216 if (hslot->count < hslot2->count)
217 goto begin;
218
219 result = udp6_lib_lookup2(net, saddr, sport,
220 daddr, hnum, dif, sdif, exact_dif,
221 hslot2, skb);
222 if (!result) {
223 unsigned int old_slot2 = slot2;
224 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
225 slot2 = hash2 & udptable->mask;
226 /* avoid searching the same slot again. */
227 if (unlikely(slot2 == old_slot2))
228 return result;
229
230 hslot2 = &udptable->hash2[slot2];
231 if (hslot->count < hslot2->count)
232 goto begin;
233
234 result = udp6_lib_lookup2(net, saddr, sport,
235 daddr, hnum, dif, sdif,
236 exact_dif, hslot2,
237 skb);
238 }
239 if (unlikely(IS_ERR(result)))
240 return NULL;
241 return result;
242 }
243 begin:
244 result = NULL;
245 badness = -1;
246 sk_for_each_rcu(sk, &hslot->head) {
247 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
248 sdif, exact_dif);
249 if (score > badness) {
250 if (sk->sk_reuseport) {
251 hash = udp6_ehashfn(net, daddr, hnum,
252 saddr, sport);
253 result = reuseport_select_sock(sk, hash, skb,
254 sizeof(struct udphdr));
255 if (unlikely(IS_ERR(result)))
256 return NULL;
257 if (result)
258 return result;
259 }
260 result = sk;
261 badness = score;
262 }
263 }
264 return result;
265 }
266 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
267
268 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
269 __be16 sport, __be16 dport,
270 struct udp_table *udptable)
271 {
272 const struct ipv6hdr *iph = ipv6_hdr(skb);
273
274 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
275 &iph->daddr, dport, inet6_iif(skb),
276 inet6_sdif(skb), udptable, skb);
277 }
278
279 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
280 __be16 sport, __be16 dport)
281 {
282 const struct ipv6hdr *iph = ipv6_hdr(skb);
283
284 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
285 &iph->daddr, dport, inet6_iif(skb),
286 inet6_sdif(skb), &udp_table, NULL);
287 }
288 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
289
290 /* Must be called under rcu_read_lock().
291 * Does increment socket refcount.
292 */
293 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
294 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
295 const struct in6_addr *daddr, __be16 dport, int dif)
296 {
297 struct sock *sk;
298
299 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
300 dif, 0, &udp_table, NULL);
301 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
302 sk = NULL;
303 return sk;
304 }
305 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
306 #endif
307
308 /* do not use the scratch area len for jumbogram: their length execeeds the
309 * scratch area space; note that the IP6CB flags is still in the first
310 * cacheline, so checking for jumbograms is cheap
311 */
312 static int udp6_skb_len(struct sk_buff *skb)
313 {
314 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
315 }
316
317 /*
318 * This should be easy, if there is something there we
319 * return it, otherwise we block.
320 */
321
322 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
323 int noblock, int flags, int *addr_len)
324 {
325 struct ipv6_pinfo *np = inet6_sk(sk);
326 struct inet_sock *inet = inet_sk(sk);
327 struct sk_buff *skb;
328 unsigned int ulen, copied;
329 int peeked, peeking, off;
330 int err;
331 int is_udplite = IS_UDPLITE(sk);
332 bool checksum_valid = false;
333 int is_udp4;
334
335 if (flags & MSG_ERRQUEUE)
336 return ipv6_recv_error(sk, msg, len, addr_len);
337
338 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
339 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
340
341 try_again:
342 peeking = flags & MSG_PEEK;
343 off = sk_peek_offset(sk, flags);
344 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
345 if (!skb)
346 return err;
347
348 ulen = udp6_skb_len(skb);
349 copied = len;
350 if (copied > ulen - off)
351 copied = ulen - off;
352 else if (copied < ulen)
353 msg->msg_flags |= MSG_TRUNC;
354
355 is_udp4 = (skb->protocol == htons(ETH_P_IP));
356
357 /*
358 * If checksum is needed at all, try to do it while copying the
359 * data. If the data is truncated, or if we only want a partial
360 * coverage checksum (UDP-Lite), do it before the copy.
361 */
362
363 if (copied < ulen || peeking ||
364 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
365 checksum_valid = udp_skb_csum_unnecessary(skb) ||
366 !__udp_lib_checksum_complete(skb);
367 if (!checksum_valid)
368 goto csum_copy_err;
369 }
370
371 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
372 if (udp_skb_is_linear(skb))
373 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
374 else
375 err = skb_copy_datagram_msg(skb, off, msg, copied);
376 } else {
377 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
378 if (err == -EINVAL)
379 goto csum_copy_err;
380 }
381 if (unlikely(err)) {
382 if (!peeked) {
383 atomic_inc(&sk->sk_drops);
384 if (is_udp4)
385 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
386 is_udplite);
387 else
388 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
389 is_udplite);
390 }
391 kfree_skb(skb);
392 return err;
393 }
394 if (!peeked) {
395 if (is_udp4)
396 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
397 is_udplite);
398 else
399 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
400 is_udplite);
401 }
402
403 sock_recv_ts_and_drops(msg, sk, skb);
404
405 /* Copy the address. */
406 if (msg->msg_name) {
407 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
408 sin6->sin6_family = AF_INET6;
409 sin6->sin6_port = udp_hdr(skb)->source;
410 sin6->sin6_flowinfo = 0;
411
412 if (is_udp4) {
413 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
414 &sin6->sin6_addr);
415 sin6->sin6_scope_id = 0;
416 } else {
417 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
418 sin6->sin6_scope_id =
419 ipv6_iface_scope_id(&sin6->sin6_addr,
420 inet6_iif(skb));
421 }
422 *addr_len = sizeof(*sin6);
423
424 if (cgroup_bpf_enabled)
425 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
426 (struct sockaddr *)sin6);
427 }
428
429 if (np->rxopt.all)
430 ip6_datagram_recv_common_ctl(sk, msg, skb);
431
432 if (is_udp4) {
433 if (inet->cmsg_flags)
434 ip_cmsg_recv_offset(msg, sk, skb,
435 sizeof(struct udphdr), off);
436 } else {
437 if (np->rxopt.all)
438 ip6_datagram_recv_specific_ctl(sk, msg, skb);
439 }
440
441 err = copied;
442 if (flags & MSG_TRUNC)
443 err = ulen;
444
445 skb_consume_udp(sk, skb, peeking ? -err : err);
446 return err;
447
448 csum_copy_err:
449 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
450 udp_skb_destructor)) {
451 if (is_udp4) {
452 UDP_INC_STATS(sock_net(sk),
453 UDP_MIB_CSUMERRORS, is_udplite);
454 UDP_INC_STATS(sock_net(sk),
455 UDP_MIB_INERRORS, is_udplite);
456 } else {
457 UDP6_INC_STATS(sock_net(sk),
458 UDP_MIB_CSUMERRORS, is_udplite);
459 UDP6_INC_STATS(sock_net(sk),
460 UDP_MIB_INERRORS, is_udplite);
461 }
462 }
463 kfree_skb(skb);
464
465 /* starting over for a new packet, but check if we need to yield */
466 cond_resched();
467 msg->msg_flags &= ~MSG_TRUNC;
468 goto try_again;
469 }
470
471 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
472 u8 type, u8 code, int offset, __be32 info,
473 struct udp_table *udptable)
474 {
475 struct ipv6_pinfo *np;
476 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
477 const struct in6_addr *saddr = &hdr->saddr;
478 const struct in6_addr *daddr = &hdr->daddr;
479 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
480 struct sock *sk;
481 int harderr;
482 int err;
483 struct net *net = dev_net(skb->dev);
484
485 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
486 inet6_iif(skb), 0, udptable, NULL);
487 if (!sk) {
488 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
489 ICMP6_MIB_INERRORS);
490 return;
491 }
492
493 harderr = icmpv6_err_convert(type, code, &err);
494 np = inet6_sk(sk);
495
496 if (type == ICMPV6_PKT_TOOBIG) {
497 if (!ip6_sk_accept_pmtu(sk))
498 goto out;
499 ip6_sk_update_pmtu(skb, sk, info);
500 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
501 harderr = 1;
502 }
503 if (type == NDISC_REDIRECT) {
504 ip6_sk_redirect(skb, sk);
505 goto out;
506 }
507
508 if (!np->recverr) {
509 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
510 goto out;
511 } else {
512 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
513 }
514
515 sk->sk_err = err;
516 sk->sk_error_report(sk);
517 out:
518 return;
519 }
520
521 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
522 {
523 int rc;
524
525 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
526 sock_rps_save_rxhash(sk, skb);
527 sk_mark_napi_id(sk, skb);
528 sk_incoming_cpu_update(sk);
529 } else {
530 sk_mark_napi_id_once(sk, skb);
531 }
532
533 rc = __udp_enqueue_schedule_skb(sk, skb);
534 if (rc < 0) {
535 int is_udplite = IS_UDPLITE(sk);
536
537 /* Note that an ENOMEM error is charged twice */
538 if (rc == -ENOMEM)
539 UDP6_INC_STATS(sock_net(sk),
540 UDP_MIB_RCVBUFERRORS, is_udplite);
541 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
542 kfree_skb(skb);
543 return -1;
544 }
545
546 return 0;
547 }
548
549 static __inline__ void udpv6_err(struct sk_buff *skb,
550 struct inet6_skb_parm *opt, u8 type,
551 u8 code, int offset, __be32 info)
552 {
553 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
554 }
555
556 static DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
557 void udpv6_encap_enable(void)
558 {
559 static_branch_enable(&udpv6_encap_needed_key);
560 }
561 EXPORT_SYMBOL(udpv6_encap_enable);
562
563 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
564 {
565 struct udp_sock *up = udp_sk(sk);
566 int is_udplite = IS_UDPLITE(sk);
567
568 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
569 goto drop;
570
571 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
572 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
573
574 /*
575 * This is an encapsulation socket so pass the skb to
576 * the socket's udp_encap_rcv() hook. Otherwise, just
577 * fall through and pass this up the UDP socket.
578 * up->encap_rcv() returns the following value:
579 * =0 if skb was successfully passed to the encap
580 * handler or was discarded by it.
581 * >0 if skb should be passed on to UDP.
582 * <0 if skb should be resubmitted as proto -N
583 */
584
585 /* if we're overly short, let UDP handle it */
586 encap_rcv = READ_ONCE(up->encap_rcv);
587 if (encap_rcv) {
588 int ret;
589
590 /* Verify checksum before giving to encap */
591 if (udp_lib_checksum_complete(skb))
592 goto csum_error;
593
594 ret = encap_rcv(sk, skb);
595 if (ret <= 0) {
596 __UDP_INC_STATS(sock_net(sk),
597 UDP_MIB_INDATAGRAMS,
598 is_udplite);
599 return -ret;
600 }
601 }
602
603 /* FALLTHROUGH -- it's a UDP Packet */
604 }
605
606 /*
607 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
608 */
609 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
610
611 if (up->pcrlen == 0) { /* full coverage was set */
612 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
613 UDP_SKB_CB(skb)->cscov, skb->len);
614 goto drop;
615 }
616 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
617 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
618 UDP_SKB_CB(skb)->cscov, up->pcrlen);
619 goto drop;
620 }
621 }
622
623 prefetch(&sk->sk_rmem_alloc);
624 if (rcu_access_pointer(sk->sk_filter) &&
625 udp_lib_checksum_complete(skb))
626 goto csum_error;
627
628 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
629 goto drop;
630
631 udp_csum_pull_header(skb);
632
633 skb_dst_drop(skb);
634
635 return __udpv6_queue_rcv_skb(sk, skb);
636
637 csum_error:
638 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
639 drop:
640 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
641 atomic_inc(&sk->sk_drops);
642 kfree_skb(skb);
643 return -1;
644 }
645
646 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
647 __be16 loc_port, const struct in6_addr *loc_addr,
648 __be16 rmt_port, const struct in6_addr *rmt_addr,
649 int dif, unsigned short hnum)
650 {
651 struct inet_sock *inet = inet_sk(sk);
652
653 if (!net_eq(sock_net(sk), net))
654 return false;
655
656 if (udp_sk(sk)->udp_port_hash != hnum ||
657 sk->sk_family != PF_INET6 ||
658 (inet->inet_dport && inet->inet_dport != rmt_port) ||
659 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
660 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
661 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
662 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
663 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
664 return false;
665 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
666 return false;
667 return true;
668 }
669
670 static void udp6_csum_zero_error(struct sk_buff *skb)
671 {
672 /* RFC 2460 section 8.1 says that we SHOULD log
673 * this error. Well, it is reasonable.
674 */
675 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
676 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
677 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
678 }
679
680 /*
681 * Note: called only from the BH handler context,
682 * so we don't need to lock the hashes.
683 */
684 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
685 const struct in6_addr *saddr, const struct in6_addr *daddr,
686 struct udp_table *udptable, int proto)
687 {
688 struct sock *sk, *first = NULL;
689 const struct udphdr *uh = udp_hdr(skb);
690 unsigned short hnum = ntohs(uh->dest);
691 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
692 unsigned int offset = offsetof(typeof(*sk), sk_node);
693 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
694 int dif = inet6_iif(skb);
695 struct hlist_node *node;
696 struct sk_buff *nskb;
697
698 if (use_hash2) {
699 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
700 udptable->mask;
701 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
702 start_lookup:
703 hslot = &udptable->hash2[hash2];
704 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
705 }
706
707 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
708 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
709 uh->source, saddr, dif, hnum))
710 continue;
711 /* If zero checksum and no_check is not on for
712 * the socket then skip it.
713 */
714 if (!uh->check && !udp_sk(sk)->no_check6_rx)
715 continue;
716 if (!first) {
717 first = sk;
718 continue;
719 }
720 nskb = skb_clone(skb, GFP_ATOMIC);
721 if (unlikely(!nskb)) {
722 atomic_inc(&sk->sk_drops);
723 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
724 IS_UDPLITE(sk));
725 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
726 IS_UDPLITE(sk));
727 continue;
728 }
729
730 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
731 consume_skb(nskb);
732 }
733
734 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
735 if (use_hash2 && hash2 != hash2_any) {
736 hash2 = hash2_any;
737 goto start_lookup;
738 }
739
740 if (first) {
741 if (udpv6_queue_rcv_skb(first, skb) > 0)
742 consume_skb(skb);
743 } else {
744 kfree_skb(skb);
745 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
746 proto == IPPROTO_UDPLITE);
747 }
748 return 0;
749 }
750
751 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
752 {
753 if (udp_sk_rx_dst_set(sk, dst)) {
754 const struct rt6_info *rt = (const struct rt6_info *)dst;
755
756 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
757 }
758 }
759
760 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
761 * return code conversion for ip layer consumption
762 */
763 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
764 struct udphdr *uh)
765 {
766 int ret;
767
768 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
769 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
770 ip6_compute_pseudo);
771
772 ret = udpv6_queue_rcv_skb(sk, skb);
773
774 /* a return value > 0 means to resubmit the input */
775 if (ret > 0)
776 return ret;
777 return 0;
778 }
779
780 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
781 int proto)
782 {
783 const struct in6_addr *saddr, *daddr;
784 struct net *net = dev_net(skb->dev);
785 struct udphdr *uh;
786 struct sock *sk;
787 u32 ulen = 0;
788
789 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
790 goto discard;
791
792 saddr = &ipv6_hdr(skb)->saddr;
793 daddr = &ipv6_hdr(skb)->daddr;
794 uh = udp_hdr(skb);
795
796 ulen = ntohs(uh->len);
797 if (ulen > skb->len)
798 goto short_packet;
799
800 if (proto == IPPROTO_UDP) {
801 /* UDP validates ulen. */
802
803 /* Check for jumbo payload */
804 if (ulen == 0)
805 ulen = skb->len;
806
807 if (ulen < sizeof(*uh))
808 goto short_packet;
809
810 if (ulen < skb->len) {
811 if (pskb_trim_rcsum(skb, ulen))
812 goto short_packet;
813 saddr = &ipv6_hdr(skb)->saddr;
814 daddr = &ipv6_hdr(skb)->daddr;
815 uh = udp_hdr(skb);
816 }
817 }
818
819 if (udp6_csum_init(skb, uh, proto))
820 goto csum_error;
821
822 /* Check if the socket is already available, e.g. due to early demux */
823 sk = skb_steal_sock(skb);
824 if (sk) {
825 struct dst_entry *dst = skb_dst(skb);
826 int ret;
827
828 if (unlikely(sk->sk_rx_dst != dst))
829 udp6_sk_rx_dst_set(sk, dst);
830
831 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
832 sock_put(sk);
833 goto report_csum_error;
834 }
835
836 ret = udp6_unicast_rcv_skb(sk, skb, uh);
837 sock_put(sk);
838 return ret;
839 }
840
841 /*
842 * Multicast receive code
843 */
844 if (ipv6_addr_is_multicast(daddr))
845 return __udp6_lib_mcast_deliver(net, skb,
846 saddr, daddr, udptable, proto);
847
848 /* Unicast */
849 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
850 if (sk) {
851 if (!uh->check && !udp_sk(sk)->no_check6_rx)
852 goto report_csum_error;
853 return udp6_unicast_rcv_skb(sk, skb, uh);
854 }
855
856 if (!uh->check)
857 goto report_csum_error;
858
859 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
860 goto discard;
861
862 if (udp_lib_checksum_complete(skb))
863 goto csum_error;
864
865 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
866 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
867
868 kfree_skb(skb);
869 return 0;
870
871 short_packet:
872 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
873 proto == IPPROTO_UDPLITE ? "-Lite" : "",
874 saddr, ntohs(uh->source),
875 ulen, skb->len,
876 daddr, ntohs(uh->dest));
877 goto discard;
878
879 report_csum_error:
880 udp6_csum_zero_error(skb);
881 csum_error:
882 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
883 discard:
884 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
885 kfree_skb(skb);
886 return 0;
887 }
888
889
890 static struct sock *__udp6_lib_demux_lookup(struct net *net,
891 __be16 loc_port, const struct in6_addr *loc_addr,
892 __be16 rmt_port, const struct in6_addr *rmt_addr,
893 int dif, int sdif)
894 {
895 unsigned short hnum = ntohs(loc_port);
896 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
897 unsigned int slot2 = hash2 & udp_table.mask;
898 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
899 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
900 struct sock *sk;
901
902 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
903 if (sk->sk_state == TCP_ESTABLISHED &&
904 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
905 return sk;
906 /* Only check first socket in chain */
907 break;
908 }
909 return NULL;
910 }
911
912 static void udp_v6_early_demux(struct sk_buff *skb)
913 {
914 struct net *net = dev_net(skb->dev);
915 const struct udphdr *uh;
916 struct sock *sk;
917 struct dst_entry *dst;
918 int dif = skb->dev->ifindex;
919 int sdif = inet6_sdif(skb);
920
921 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
922 sizeof(struct udphdr)))
923 return;
924
925 uh = udp_hdr(skb);
926
927 if (skb->pkt_type == PACKET_HOST)
928 sk = __udp6_lib_demux_lookup(net, uh->dest,
929 &ipv6_hdr(skb)->daddr,
930 uh->source, &ipv6_hdr(skb)->saddr,
931 dif, sdif);
932 else
933 return;
934
935 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
936 return;
937
938 skb->sk = sk;
939 skb->destructor = sock_efree;
940 dst = READ_ONCE(sk->sk_rx_dst);
941
942 if (dst)
943 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
944 if (dst) {
945 /* set noref for now.
946 * any place which wants to hold dst has to call
947 * dst_hold_safe()
948 */
949 skb_dst_set_noref(skb, dst);
950 }
951 }
952
953 static __inline__ int udpv6_rcv(struct sk_buff *skb)
954 {
955 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
956 }
957
958 /*
959 * Throw away all pending data and cancel the corking. Socket is locked.
960 */
961 static void udp_v6_flush_pending_frames(struct sock *sk)
962 {
963 struct udp_sock *up = udp_sk(sk);
964
965 if (up->pending == AF_INET)
966 udp_flush_pending_frames(sk);
967 else if (up->pending) {
968 up->len = 0;
969 up->pending = 0;
970 ip6_flush_pending_frames(sk);
971 }
972 }
973
974 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
975 int addr_len)
976 {
977 /* The following checks are replicated from __ip6_datagram_connect()
978 * and intended to prevent BPF program called below from accessing
979 * bytes that are out of the bound specified by user in addr_len.
980 */
981 if (uaddr->sa_family == AF_INET) {
982 if (__ipv6_only_sock(sk))
983 return -EAFNOSUPPORT;
984 return udp_pre_connect(sk, uaddr, addr_len);
985 }
986
987 if (addr_len < SIN6_LEN_RFC2133)
988 return -EINVAL;
989
990 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
991 }
992
993 /**
994 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
995 * @sk: socket we are sending on
996 * @skb: sk_buff containing the filled-in UDP header
997 * (checksum field must be zeroed out)
998 */
999 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1000 const struct in6_addr *saddr,
1001 const struct in6_addr *daddr, int len)
1002 {
1003 unsigned int offset;
1004 struct udphdr *uh = udp_hdr(skb);
1005 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1006 __wsum csum = 0;
1007
1008 if (!frags) {
1009 /* Only one fragment on the socket. */
1010 skb->csum_start = skb_transport_header(skb) - skb->head;
1011 skb->csum_offset = offsetof(struct udphdr, check);
1012 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1013 } else {
1014 /*
1015 * HW-checksum won't work as there are two or more
1016 * fragments on the socket so that all csums of sk_buffs
1017 * should be together
1018 */
1019 offset = skb_transport_offset(skb);
1020 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1021 csum = skb->csum;
1022
1023 skb->ip_summed = CHECKSUM_NONE;
1024
1025 do {
1026 csum = csum_add(csum, frags->csum);
1027 } while ((frags = frags->next));
1028
1029 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1030 csum);
1031 if (uh->check == 0)
1032 uh->check = CSUM_MANGLED_0;
1033 }
1034 }
1035
1036 /*
1037 * Sending
1038 */
1039
1040 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1041 struct inet_cork *cork)
1042 {
1043 struct sock *sk = skb->sk;
1044 struct udphdr *uh;
1045 int err = 0;
1046 int is_udplite = IS_UDPLITE(sk);
1047 __wsum csum = 0;
1048 int offset = skb_transport_offset(skb);
1049 int len = skb->len - offset;
1050 int datalen = len - sizeof(*uh);
1051
1052 /*
1053 * Create a UDP header
1054 */
1055 uh = udp_hdr(skb);
1056 uh->source = fl6->fl6_sport;
1057 uh->dest = fl6->fl6_dport;
1058 uh->len = htons(len);
1059 uh->check = 0;
1060
1061 if (cork->gso_size) {
1062 const int hlen = skb_network_header_len(skb) +
1063 sizeof(struct udphdr);
1064
1065 if (hlen + cork->gso_size > cork->fragsize) {
1066 kfree_skb(skb);
1067 return -EINVAL;
1068 }
1069 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
1070 kfree_skb(skb);
1071 return -EINVAL;
1072 }
1073 if (udp_sk(sk)->no_check6_tx) {
1074 kfree_skb(skb);
1075 return -EINVAL;
1076 }
1077 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1078 dst_xfrm(skb_dst(skb))) {
1079 kfree_skb(skb);
1080 return -EIO;
1081 }
1082
1083 if (datalen > cork->gso_size) {
1084 skb_shinfo(skb)->gso_size = cork->gso_size;
1085 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1086 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1087 cork->gso_size);
1088 }
1089 goto csum_partial;
1090 }
1091
1092 if (is_udplite)
1093 csum = udplite_csum(skb);
1094 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1095 skb->ip_summed = CHECKSUM_NONE;
1096 goto send;
1097 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1098 csum_partial:
1099 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1100 goto send;
1101 } else
1102 csum = udp_csum(skb);
1103
1104 /* add protocol-dependent pseudo-header */
1105 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1106 len, fl6->flowi6_proto, csum);
1107 if (uh->check == 0)
1108 uh->check = CSUM_MANGLED_0;
1109
1110 send:
1111 err = ip6_send_skb(skb);
1112 if (err) {
1113 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1114 UDP6_INC_STATS(sock_net(sk),
1115 UDP_MIB_SNDBUFERRORS, is_udplite);
1116 err = 0;
1117 }
1118 } else {
1119 UDP6_INC_STATS(sock_net(sk),
1120 UDP_MIB_OUTDATAGRAMS, is_udplite);
1121 }
1122 return err;
1123 }
1124
1125 static int udp_v6_push_pending_frames(struct sock *sk)
1126 {
1127 struct sk_buff *skb;
1128 struct udp_sock *up = udp_sk(sk);
1129 struct flowi6 fl6;
1130 int err = 0;
1131
1132 if (up->pending == AF_INET)
1133 return udp_push_pending_frames(sk);
1134
1135 /* ip6_finish_skb will release the cork, so make a copy of
1136 * fl6 here.
1137 */
1138 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1139
1140 skb = ip6_finish_skb(sk);
1141 if (!skb)
1142 goto out;
1143
1144 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1145
1146 out:
1147 up->len = 0;
1148 up->pending = 0;
1149 return err;
1150 }
1151
1152 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1153 {
1154 struct ipv6_txoptions opt_space;
1155 struct udp_sock *up = udp_sk(sk);
1156 struct inet_sock *inet = inet_sk(sk);
1157 struct ipv6_pinfo *np = inet6_sk(sk);
1158 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1159 struct in6_addr *daddr, *final_p, final;
1160 struct ipv6_txoptions *opt = NULL;
1161 struct ipv6_txoptions *opt_to_free = NULL;
1162 struct ip6_flowlabel *flowlabel = NULL;
1163 struct flowi6 fl6;
1164 struct dst_entry *dst;
1165 struct ipcm6_cookie ipc6;
1166 int addr_len = msg->msg_namelen;
1167 bool connected = false;
1168 int ulen = len;
1169 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1170 int err;
1171 int is_udplite = IS_UDPLITE(sk);
1172 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1173
1174 ipcm6_init(&ipc6);
1175 ipc6.gso_size = up->gso_size;
1176 ipc6.sockc.tsflags = sk->sk_tsflags;
1177
1178 /* destination address check */
1179 if (sin6) {
1180 if (addr_len < offsetof(struct sockaddr, sa_data))
1181 return -EINVAL;
1182
1183 switch (sin6->sin6_family) {
1184 case AF_INET6:
1185 if (addr_len < SIN6_LEN_RFC2133)
1186 return -EINVAL;
1187 daddr = &sin6->sin6_addr;
1188 if (ipv6_addr_any(daddr) &&
1189 ipv6_addr_v4mapped(&np->saddr))
1190 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1191 daddr);
1192 break;
1193 case AF_INET:
1194 goto do_udp_sendmsg;
1195 case AF_UNSPEC:
1196 msg->msg_name = sin6 = NULL;
1197 msg->msg_namelen = addr_len = 0;
1198 daddr = NULL;
1199 break;
1200 default:
1201 return -EINVAL;
1202 }
1203 } else if (!up->pending) {
1204 if (sk->sk_state != TCP_ESTABLISHED)
1205 return -EDESTADDRREQ;
1206 daddr = &sk->sk_v6_daddr;
1207 } else
1208 daddr = NULL;
1209
1210 if (daddr) {
1211 if (ipv6_addr_v4mapped(daddr)) {
1212 struct sockaddr_in sin;
1213 sin.sin_family = AF_INET;
1214 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1215 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1216 msg->msg_name = &sin;
1217 msg->msg_namelen = sizeof(sin);
1218 do_udp_sendmsg:
1219 if (__ipv6_only_sock(sk))
1220 return -ENETUNREACH;
1221 return udp_sendmsg(sk, msg, len);
1222 }
1223 }
1224
1225 if (up->pending == AF_INET)
1226 return udp_sendmsg(sk, msg, len);
1227
1228 /* Rough check on arithmetic overflow,
1229 better check is made in ip6_append_data().
1230 */
1231 if (len > INT_MAX - sizeof(struct udphdr))
1232 return -EMSGSIZE;
1233
1234 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1235 if (up->pending) {
1236 /*
1237 * There are pending frames.
1238 * The socket lock must be held while it's corked.
1239 */
1240 lock_sock(sk);
1241 if (likely(up->pending)) {
1242 if (unlikely(up->pending != AF_INET6)) {
1243 release_sock(sk);
1244 return -EAFNOSUPPORT;
1245 }
1246 dst = NULL;
1247 goto do_append_data;
1248 }
1249 release_sock(sk);
1250 }
1251 ulen += sizeof(struct udphdr);
1252
1253 memset(&fl6, 0, sizeof(fl6));
1254
1255 if (sin6) {
1256 if (sin6->sin6_port == 0)
1257 return -EINVAL;
1258
1259 fl6.fl6_dport = sin6->sin6_port;
1260 daddr = &sin6->sin6_addr;
1261
1262 if (np->sndflow) {
1263 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1264 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1265 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1266 if (!flowlabel)
1267 return -EINVAL;
1268 }
1269 }
1270
1271 /*
1272 * Otherwise it will be difficult to maintain
1273 * sk->sk_dst_cache.
1274 */
1275 if (sk->sk_state == TCP_ESTABLISHED &&
1276 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1277 daddr = &sk->sk_v6_daddr;
1278
1279 if (addr_len >= sizeof(struct sockaddr_in6) &&
1280 sin6->sin6_scope_id &&
1281 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1282 fl6.flowi6_oif = sin6->sin6_scope_id;
1283 } else {
1284 if (sk->sk_state != TCP_ESTABLISHED)
1285 return -EDESTADDRREQ;
1286
1287 fl6.fl6_dport = inet->inet_dport;
1288 daddr = &sk->sk_v6_daddr;
1289 fl6.flowlabel = np->flow_label;
1290 connected = true;
1291 }
1292
1293 if (!fl6.flowi6_oif)
1294 fl6.flowi6_oif = sk->sk_bound_dev_if;
1295
1296 if (!fl6.flowi6_oif)
1297 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1298
1299 fl6.flowi6_mark = sk->sk_mark;
1300 fl6.flowi6_uid = sk->sk_uid;
1301
1302 if (msg->msg_controllen) {
1303 opt = &opt_space;
1304 memset(opt, 0, sizeof(struct ipv6_txoptions));
1305 opt->tot_len = sizeof(*opt);
1306 ipc6.opt = opt;
1307
1308 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1309 if (err > 0)
1310 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1311 &ipc6);
1312 if (err < 0) {
1313 fl6_sock_release(flowlabel);
1314 return err;
1315 }
1316 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1317 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1318 if (!flowlabel)
1319 return -EINVAL;
1320 }
1321 if (!(opt->opt_nflen|opt->opt_flen))
1322 opt = NULL;
1323 connected = false;
1324 }
1325 if (!opt) {
1326 opt = txopt_get(np);
1327 opt_to_free = opt;
1328 }
1329 if (flowlabel)
1330 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1331 opt = ipv6_fixup_options(&opt_space, opt);
1332 ipc6.opt = opt;
1333
1334 fl6.flowi6_proto = sk->sk_protocol;
1335 fl6.daddr = *daddr;
1336 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1337 fl6.saddr = np->saddr;
1338 fl6.fl6_sport = inet->inet_sport;
1339
1340 if (cgroup_bpf_enabled && !connected) {
1341 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1342 (struct sockaddr *)sin6, &fl6.saddr);
1343 if (err)
1344 goto out_no_dst;
1345 if (sin6) {
1346 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1347 /* BPF program rewrote IPv6-only by IPv4-mapped
1348 * IPv6. It's currently unsupported.
1349 */
1350 err = -ENOTSUPP;
1351 goto out_no_dst;
1352 }
1353 if (sin6->sin6_port == 0) {
1354 /* BPF program set invalid port. Reject it. */
1355 err = -EINVAL;
1356 goto out_no_dst;
1357 }
1358 fl6.fl6_dport = sin6->sin6_port;
1359 fl6.daddr = sin6->sin6_addr;
1360 }
1361 }
1362
1363 if (ipv6_addr_any(&fl6.daddr))
1364 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1365
1366 final_p = fl6_update_dst(&fl6, opt, &final);
1367 if (final_p)
1368 connected = false;
1369
1370 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1371 fl6.flowi6_oif = np->mcast_oif;
1372 connected = false;
1373 } else if (!fl6.flowi6_oif)
1374 fl6.flowi6_oif = np->ucast_oif;
1375
1376 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1377
1378 if (ipc6.tclass < 0)
1379 ipc6.tclass = np->tclass;
1380
1381 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1382
1383 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1384 if (IS_ERR(dst)) {
1385 err = PTR_ERR(dst);
1386 dst = NULL;
1387 goto out;
1388 }
1389
1390 if (ipc6.hlimit < 0)
1391 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1392
1393 if (msg->msg_flags&MSG_CONFIRM)
1394 goto do_confirm;
1395 back_from_confirm:
1396
1397 /* Lockless fast path for the non-corking case */
1398 if (!corkreq) {
1399 struct inet_cork_full cork;
1400 struct sk_buff *skb;
1401
1402 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1403 sizeof(struct udphdr), &ipc6,
1404 &fl6, (struct rt6_info *)dst,
1405 msg->msg_flags, &cork);
1406 err = PTR_ERR(skb);
1407 if (!IS_ERR_OR_NULL(skb))
1408 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1409 goto out;
1410 }
1411
1412 lock_sock(sk);
1413 if (unlikely(up->pending)) {
1414 /* The socket is already corked while preparing it. */
1415 /* ... which is an evident application bug. --ANK */
1416 release_sock(sk);
1417
1418 net_dbg_ratelimited("udp cork app bug 2\n");
1419 err = -EINVAL;
1420 goto out;
1421 }
1422
1423 up->pending = AF_INET6;
1424
1425 do_append_data:
1426 if (ipc6.dontfrag < 0)
1427 ipc6.dontfrag = np->dontfrag;
1428 up->len += ulen;
1429 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1430 &ipc6, &fl6, (struct rt6_info *)dst,
1431 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1432 if (err)
1433 udp_v6_flush_pending_frames(sk);
1434 else if (!corkreq)
1435 err = udp_v6_push_pending_frames(sk);
1436 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1437 up->pending = 0;
1438
1439 if (err > 0)
1440 err = np->recverr ? net_xmit_errno(err) : 0;
1441 release_sock(sk);
1442
1443 out:
1444 dst_release(dst);
1445 out_no_dst:
1446 fl6_sock_release(flowlabel);
1447 txopt_put(opt_to_free);
1448 if (!err)
1449 return len;
1450 /*
1451 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1452 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1453 * we don't have a good statistic (IpOutDiscards but it can be too many
1454 * things). We could add another new stat but at least for now that
1455 * seems like overkill.
1456 */
1457 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1458 UDP6_INC_STATS(sock_net(sk),
1459 UDP_MIB_SNDBUFERRORS, is_udplite);
1460 }
1461 return err;
1462
1463 do_confirm:
1464 if (msg->msg_flags & MSG_PROBE)
1465 dst_confirm_neigh(dst, &fl6.daddr);
1466 if (!(msg->msg_flags&MSG_PROBE) || len)
1467 goto back_from_confirm;
1468 err = 0;
1469 goto out;
1470 }
1471
1472 void udpv6_destroy_sock(struct sock *sk)
1473 {
1474 struct udp_sock *up = udp_sk(sk);
1475 lock_sock(sk);
1476 udp_v6_flush_pending_frames(sk);
1477 release_sock(sk);
1478
1479 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
1480 void (*encap_destroy)(struct sock *sk);
1481 encap_destroy = READ_ONCE(up->encap_destroy);
1482 if (encap_destroy)
1483 encap_destroy(sk);
1484 }
1485
1486 inet6_destroy_sock(sk);
1487 }
1488
1489 /*
1490 * Socket option code for UDP
1491 */
1492 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1493 char __user *optval, unsigned int optlen)
1494 {
1495 if (level == SOL_UDP || level == SOL_UDPLITE)
1496 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1497 udp_v6_push_pending_frames);
1498 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1499 }
1500
1501 #ifdef CONFIG_COMPAT
1502 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1503 char __user *optval, unsigned int optlen)
1504 {
1505 if (level == SOL_UDP || level == SOL_UDPLITE)
1506 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1507 udp_v6_push_pending_frames);
1508 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1509 }
1510 #endif
1511
1512 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1513 char __user *optval, int __user *optlen)
1514 {
1515 if (level == SOL_UDP || level == SOL_UDPLITE)
1516 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1517 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1518 }
1519
1520 #ifdef CONFIG_COMPAT
1521 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1522 char __user *optval, int __user *optlen)
1523 {
1524 if (level == SOL_UDP || level == SOL_UDPLITE)
1525 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1526 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1527 }
1528 #endif
1529
1530 /* thinking of making this const? Don't.
1531 * early_demux can change based on sysctl.
1532 */
1533 static struct inet6_protocol udpv6_protocol = {
1534 .early_demux = udp_v6_early_demux,
1535 .early_demux_handler = udp_v6_early_demux,
1536 .handler = udpv6_rcv,
1537 .err_handler = udpv6_err,
1538 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1539 };
1540
1541 /* ------------------------------------------------------------------------ */
1542 #ifdef CONFIG_PROC_FS
1543 int udp6_seq_show(struct seq_file *seq, void *v)
1544 {
1545 if (v == SEQ_START_TOKEN) {
1546 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1547 } else {
1548 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1549 struct inet_sock *inet = inet_sk(v);
1550 __u16 srcp = ntohs(inet->inet_sport);
1551 __u16 destp = ntohs(inet->inet_dport);
1552 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1553 udp_rqueue_get(v), bucket);
1554 }
1555 return 0;
1556 }
1557
1558 const struct seq_operations udp6_seq_ops = {
1559 .start = udp_seq_start,
1560 .next = udp_seq_next,
1561 .stop = udp_seq_stop,
1562 .show = udp6_seq_show,
1563 };
1564 EXPORT_SYMBOL(udp6_seq_ops);
1565
1566 static struct udp_seq_afinfo udp6_seq_afinfo = {
1567 .family = AF_INET6,
1568 .udp_table = &udp_table,
1569 };
1570
1571 int __net_init udp6_proc_init(struct net *net)
1572 {
1573 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1574 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1575 return -ENOMEM;
1576 return 0;
1577 }
1578
1579 void udp6_proc_exit(struct net *net)
1580 {
1581 remove_proc_entry("udp6", net->proc_net);
1582 }
1583 #endif /* CONFIG_PROC_FS */
1584
1585 /* ------------------------------------------------------------------------ */
1586
1587 struct proto udpv6_prot = {
1588 .name = "UDPv6",
1589 .owner = THIS_MODULE,
1590 .close = udp_lib_close,
1591 .pre_connect = udpv6_pre_connect,
1592 .connect = ip6_datagram_connect,
1593 .disconnect = udp_disconnect,
1594 .ioctl = udp_ioctl,
1595 .init = udp_init_sock,
1596 .destroy = udpv6_destroy_sock,
1597 .setsockopt = udpv6_setsockopt,
1598 .getsockopt = udpv6_getsockopt,
1599 .sendmsg = udpv6_sendmsg,
1600 .recvmsg = udpv6_recvmsg,
1601 .release_cb = ip6_datagram_release_cb,
1602 .hash = udp_lib_hash,
1603 .unhash = udp_lib_unhash,
1604 .rehash = udp_v6_rehash,
1605 .get_port = udp_v6_get_port,
1606 .memory_allocated = &udp_memory_allocated,
1607 .sysctl_mem = sysctl_udp_mem,
1608 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1609 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1610 .obj_size = sizeof(struct udp6_sock),
1611 .h.udp_table = &udp_table,
1612 #ifdef CONFIG_COMPAT
1613 .compat_setsockopt = compat_udpv6_setsockopt,
1614 .compat_getsockopt = compat_udpv6_getsockopt,
1615 #endif
1616 .diag_destroy = udp_abort,
1617 };
1618
1619 static struct inet_protosw udpv6_protosw = {
1620 .type = SOCK_DGRAM,
1621 .protocol = IPPROTO_UDP,
1622 .prot = &udpv6_prot,
1623 .ops = &inet6_dgram_ops,
1624 .flags = INET_PROTOSW_PERMANENT,
1625 };
1626
1627 int __init udpv6_init(void)
1628 {
1629 int ret;
1630
1631 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1632 if (ret)
1633 goto out;
1634
1635 ret = inet6_register_protosw(&udpv6_protosw);
1636 if (ret)
1637 goto out_udpv6_protocol;
1638 out:
1639 return ret;
1640
1641 out_udpv6_protocol:
1642 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1643 goto out;
1644 }
1645
1646 void udpv6_exit(void)
1647 {
1648 inet6_unregister_protosw(&udpv6_protosw);
1649 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1650 }
Linux with added FPC-III support